*From: *Christopher Baum <baum@bc.edu <mailto:baum@bc.edu>>
*Date: *May 3, 2007 6:40:35 AM EDT
*To: *statalist@hsphsun2.harvard.edu
<mailto:statalist@hsphsun2.harvard.edu>
*Cc: *vlw@stata.com <mailto:vlw@stata.com>
*Subject: **sureg*
sacrificial lamb
There are numerous cases where -sureg- run on several equations is
equivalent in every way to running OLS on each equation. -sureg (y x1
x2 x3)- is one of them. It is doing nothing at all beyond OLS, except
for reporting a large-sample VCE.
That said, -sureg- appears to have a bug. If you do
webuse auto
reg price mpg headroom trunk
sureg (price mpg headroom trunk)
sureg (price mpg headroom trunk), small
The latter sureg results should equal OLS in terms of the standard
errors and the t-stats: that's the meaning of -small-. But that
option appears to do nothing at all.
Compare with
ivreg2 price mpg headroom trunk
ivreg2 price mpg headroom trunk, small
The first will reproduce -sureg- (with or without the small option);
the second will reproduce -regress-. So there appears to be something
wrong with -sureg-'s small option; it labels things as t-stats but
does not appear to calculate the VCE with the smaller denominator
that -small- implies. This problem applies even if -sureg- is used as
intended, e.g. to estimate multiple equations.
One more for the old cert script, developers?
Kit
On May 2, 2007, at 2:33 AM, statalist-digest wrote:
It seems like sureg command works even with one equation (sureg y x).
I was not aware that Zellner's SUR can work with only one equation,
rather than two, therefore I assume that one-equation case corrects for
the correlated errors by using an instrument. Am I right? If so, which
instrument is it using? I would appreciate it if anyone leads me to a
more technical discussion, or explains what the estimation command is
doing.